On the evidence for clustering in the arrival directions of AGASA's ultrahigh energy cosmic rays

Previous analyses of cosmic rays above 4 x 10(19) eV observed by the AGASA experiment have suggested that their arrival directions may be clustered. However, estimates of the chance probability of this clustering signal vary from 10(-2) to 10(-6) and beyond. It is essential that the strength of this evidence be well understood in order to compare it with anisotropy studies in other cosmic ray experiments. We apply two methods for extracting a meaningful significance from this data set: one can scan for the cuts which optimize the clustering signal, using simulations to determine the appropriate statistical penalty for the scan. This analysis finds a chance probability of about 0.3%. Alternatively, one can optimize the cuts with a first set of data, and then apply them to the remaining data directly without statistical penalty. One can extend the statistical power of this test by considering cross-correlation between the initial data and the remaining data, as long as the initial clustering signal is not included. While the scan is more useful in general, in the present case only splitting the data set offers an unbiased test of the clustering hypothesis. Using this test we find that the AGASA data is consistent at the 8% level with the null hypothesis of isotropically distributed arrival directions. (C) 2004 Elsevier B.V. All rights reserved.